Microstructure-based Simulation And Optimization Methods For Polymerization Processes

Polymers are widely applied in our daily life according to their various performances; they could be found in automobiles, bottles, films, and electric appliances. The end-use quality of polymers is directly determined by their corresponding microstructures that can be very different for different grades. According to the number of involved monomers, the polymerization processes could be simply classified into homopolymerization and copolymerization. In homopolymerization, molecular weight distribution (MWD) is the most important microstructure of polymer. In copolymerization, the microstructural indexes include molecular weight distribution, chemical composition distribution (CCD), sequence length distribution (SLD), and sometimes even short chain branching and long chain branching. To improve the productivity and the performance of polymer, how to solve those complex computing problems based on those microstructures better and faster is still a challenging task.In this paper, a series of microstructural simulation and optimization methods are proposed to improve the corresponding computing efficiency. The detailed research work can be summarized as follows:1. Parallel simulation method in homopolymerization. The conventional method for the dynamic simulation of MWD is accomplished by conducting the three modules, including moment method, instantaneous Flory method, and cumulative Flory method, in a serial mode. In this project, the primary serial computation is divided and reconstructed using multiple threads by analyzing the relationships and communications among the moment method and the MWD calculation of each active site. A parallel strategy is proposed to accelerate the computation with a maximum speedup ratio and a corresponding decomposition algorithm is proposed to select the least number of parallel cores.Finally, the grade transition for a slurry process involving high-density polyethylene (HDPE) is simulated.2. Optimization method of grade transition in homopolymerization. The conventional methods for dynamic optimization of grade transition process are based on the average molecular weight or the molecular weight distribution. A novel strategy for MWD dynamic optimization is proposed based on a theoretical analysis of the relationships among state variables. A theorem and case validation are further provided to show that the real-time MWD will cease to vary after the stabilization of the selected reactor state variables in the moment model. Thus, dynamic MWD optimization can be conducted by solving a small-scale dynamic moment model combined with a steady state MWD model. Finally, the dynamic optimization of grade transition for an HDPE slurry process is presented with the proposed method.3. Parallel simulation method in copolymerization. From homopolymerization to copolymerization, there is not only the change of kinetic mechanism, but also the representation of microstructures. Due to the more complex microstructures, the Monte Carlo method is suitable to simulate the microstructural distributions. However, this method is inefficient from the point of view of computation time. A parallel method is developed in this project for the Monte Carlo simulation on a graphics processing unit platform. Both steady state and dynamic state cases are presented to show the accuracy and efficiency of the proposed method. The computation time of the proposed method is greatly decreased by at least 30-fold compared with the time required on CPU platform.4. Uncertainty optimization method in copolymerization. The parallel Monte Carlo method could greatly accelerate the simulation, which makes the optimization of the microstructural distribution with the Monte Carlo simulation feasible. Since the Monte Carlo simulation includes stochastic results, it is impractical to obtain derivatives. Hence, the derivative-free optimization method is preferred. Also due to the stochastic feature in the Monte Carlo simulation, the uncertainty has to be considered in the optimization. As the number of simulated chains of the Monte Carlo simulation directly affects the simulation results, an efficient way to choose the number of simulated chains for the optimization is proposed to reduce the time cost of the computation. Finally, a ternary polymerization case is presented to show the efficiency of the proposed method.Finally, a summary of the research work is given, and the perspectives of the future work are presented.